Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A system comprising a computer including a processor and a memory, the memory including instructions such that the processor is programmed to: receive a synthetic image at a first deep neural network; and determine, via the first deep neural network, a prediction indicative of whether the synthetic image is machine-generated or is sourced from a real data distribution; wherein the prediction is used as feedback to a generator included in the first deep neural network to train the generator and then communicated from the generator to a discriminator included in the first deep neural network to update the discriminator training; and wherein the prediction comprises a quantitative measure of photorealism of synthetic image.
The system operates in the domain of deep learning and synthetic image generation, addressing the challenge of distinguishing between machine-generated synthetic images and real-world images. It employs a deep neural network to evaluate the photorealism of synthetic images, providing feedback to improve the generator and discriminator components of the network. The system includes a computer with a processor and memory, where the memory stores instructions for the processor to execute. The processor receives a synthetic image and processes it through a first deep neural network to determine whether the image is machine-generated or sourced from a real data distribution. The prediction output includes a quantitative measure of photorealism, which is used as feedback to train the generator. This feedback is then communicated to the discriminator to update its training. The generator and discriminator work together in an adversarial manner, where the generator aims to produce increasingly realistic images, while the discriminator improves its ability to distinguish between real and synthetic images. This iterative process enhances the system's ability to generate highly photorealistic synthetic images and accurately assess their authenticity. The system is particularly useful in applications requiring high-quality synthetic image generation, such as data augmentation, virtual reality, and deepfake detection.
2. The system of claim 1 , wherein the quantitative measure of the photorealism indicates how close the synthetic image corresponds to an image from the real data distribution.
The invention relates to systems for evaluating photorealism in synthetic images, addressing the challenge of assessing how closely generated images resemble real-world images. The system includes a photorealism assessment module that computes a quantitative measure of photorealism for a synthetic image by comparing it to a real data distribution. This measure indicates the degree of similarity between the synthetic image and real images, helping to determine the quality and authenticity of the generated content. The system may also include a synthetic image generation module that produces the synthetic image using techniques such as generative adversarial networks (GANs) or other deep learning methods. The photorealism assessment module analyzes the synthetic image by evaluating features like texture, lighting, and structural coherence, then compares these features to those of real images in a reference dataset. The quantitative measure can be used to refine the generation process, ensuring that subsequent synthetic images more closely match real-world visual characteristics. This approach improves the realism of synthetic images, making them more suitable for applications in virtual reality, gaming, and media production.
3. The system of claim 1 , wherein the synthetic image depicts a plurality of objects.
The invention relates to a system for generating and analyzing synthetic images, particularly for applications in computer vision, machine learning, or augmented reality. The system addresses the challenge of creating realistic and diverse synthetic images that can be used to train or test algorithms without relying solely on real-world data, which may be limited or expensive to obtain. The system generates synthetic images that depict multiple objects, each with configurable properties such as shape, size, color, texture, and spatial arrangement. These objects can be static or dynamic, and their interactions can be simulated to create realistic scenes. The system may also include features for adjusting lighting conditions, backgrounds, and environmental factors to enhance realism. Additionally, the system can incorporate metadata or annotations to describe the objects and their properties, enabling automated analysis or further processing. The synthetic images can be used for training machine learning models, validating algorithms, or simulating real-world scenarios in fields such as autonomous vehicles, robotics, or virtual reality. By generating diverse and controlled synthetic datasets, the system helps improve the robustness and accuracy of computer vision systems while reducing dependency on real-world data collection.
4. The system of claim 3 , wherein the synthetic image depicts the plurality of objects corresponding to an image view of a simulated image.
This invention relates to systems for generating synthetic images of objects in a simulated environment. The system addresses the challenge of creating realistic visual representations of multiple objects from different perspectives, which is critical for applications such as virtual reality, augmented reality, and computer vision training. The system generates a synthetic image that accurately depicts a plurality of objects as they would appear in a simulated image view. This involves simulating the visual characteristics of the objects, including their shapes, textures, and spatial relationships, to produce a coherent and realistic representation. The synthetic image is generated based on predefined or dynamically adjusted parameters, ensuring that the objects are rendered in a manner consistent with the simulated environment. The system may also incorporate additional features, such as lighting effects, shadows, and depth perception, to enhance the realism of the synthetic image. By providing a simulated image view, the system enables applications that require accurate visualizations of objects in virtual or augmented environments. This technology is particularly useful in scenarios where real-world imaging is impractical or costly, such as in training simulations, virtual prototyping, and autonomous system development. The system ensures that the synthetic image maintains fidelity to the simulated conditions, making it a valuable tool for testing and validation in various technical domains.
5. The system of claim 4 , wherein the simulated image is generated by an image generator.
The invention relates to a system for generating simulated images, particularly in the context of computer vision, machine learning, or augmented reality applications. The system addresses the challenge of creating realistic or synthetic images for training models, testing algorithms, or enhancing user experiences in virtual environments. The system includes an image generator that produces simulated images based on input data, such as parameters, templates, or real-world data. The generated images are designed to mimic real-world scenes, objects, or conditions, enabling applications like autonomous vehicle training, virtual prototyping, or augmented reality simulations. The image generator may use techniques such as generative adversarial networks (GANs), neural rendering, or procedural generation to produce high-fidelity images. The system ensures that the simulated images are visually coherent and contextually relevant, improving the accuracy and reliability of downstream applications. The invention enhances the ability to create large datasets of synthetic images, reducing the need for expensive or time-consuming real-world data collection.
6. The system of claim 5 , wherein the image generator comprises a gaming engine.
A system for generating and displaying images in a gaming environment addresses the need for high-quality, interactive visual content in real-time applications. The system includes an image generator that creates visual content based on input data, such as user interactions or environmental inputs. The image generator utilizes a gaming engine, which is a software framework designed for rendering graphics, simulating physics, and managing game logic. This engine processes input data to produce dynamic, responsive images that can be displayed on a screen or other output device. The system may also include additional components, such as input devices for capturing user actions or sensors for detecting environmental conditions, which feed data into the image generator. The gaming engine within the image generator ensures that the generated images are visually consistent, interactive, and optimized for performance, making it suitable for applications like virtual reality, augmented reality, or interactive simulations. The system enhances user engagement by providing immersive, real-time visual feedback based on dynamic inputs.
7. The system of claim 4 , wherein the synthetic image is generated via a second deep neural network based on the simulated image.
A system for generating synthetic images using deep neural networks addresses the challenge of creating realistic or augmented visual data for applications such as training machine learning models, computer vision tasks, or virtual simulations. The system leverages a first deep neural network to produce a simulated image from input data, which may include real-world images, sensor data, or other relevant inputs. This simulated image serves as a foundation for further processing. A second deep neural network then refines or transforms the simulated image into a synthetic image, enhancing its realism, detail, or specific features. The second network may incorporate additional data, such as environmental conditions, object attributes, or user-defined parameters, to tailor the output. This two-stage approach allows for greater flexibility and control over the final synthetic image, enabling applications in fields like autonomous driving, medical imaging, or augmented reality. The system improves upon traditional methods by using deep learning to automate and optimize the generation of high-quality synthetic images, reducing the need for manual intervention or extensive real-world data collection.
8. The system of claim 7 , wherein the second deep neural network comprises an encoder-decoder neural network and the first deep neural network comprises a convolutional neural network.
9. The system of claim 1 , wherein the processor is further programmed to: determine that the synthetic image is sourced from the real data distribution when the prediction is greater than or equal to a predetermined threshold.
10. The system of claim 9 , wherein the processor is further programmed to: store the synthetic image in a database when the prediction is greater than or equal to the predetermined threshold.
11. A method comprising: receiving a synthetic image at a first deep neural network; and determining, via the first deep neural network, a prediction indicative of whether the synthetic image is machine-generated or is sourced from a real data distribution; wherein the prediction is used as feedback to a generator included in the first deep neural network to train the generator and then communicated from the generator to a discriminator included in the first deep neural network to update the discriminator training; and wherein the prediction comprises a quantitative measure of photorealism of synthetic image.
12. The method of claim 11 , wherein the quantitative measure of the photorealism indicates how close the synthetic image corresponds to an image from the real data distribution.
This invention relates to evaluating photorealism in synthetic images, particularly in machine learning applications where synthetic data is generated to mimic real-world images. The problem addressed is the need to assess how closely synthetic images resemble real images, ensuring they are indistinguishable from real data for training or testing purposes. The method involves generating a synthetic image using a generative model, such as a generative adversarial network (GAN) or a variational autoencoder (VAE). The synthetic image is then compared to a reference image from a real data distribution to compute a quantitative measure of photorealism. This measure indicates the degree of similarity between the synthetic image and real images, helping to refine the generative model for improved realism. The comparison may involve statistical analysis, perceptual metrics, or deep learning-based evaluation techniques. The goal is to ensure synthetic images are sufficiently realistic for applications like data augmentation, privacy-preserving data generation, or training machine learning models. The method may also include iterative refinement of the generative model based on the computed photorealism measure to enhance the quality of future synthetic images.
13. The method of claim 11 , wherein the synthetic image depicts a plurality of objects.
14. The method of claim 13 , wherein the synthetic image depicts the plurality of objects corresponding to an image view of a simulated image.
This invention relates to generating synthetic images for training or evaluating machine learning models, particularly in computer vision applications. The problem addressed is the need for realistic synthetic training data that accurately represents real-world scenarios, including multiple objects in a simulated environment. The method involves creating a synthetic image that depicts a plurality of objects, where these objects correspond to an image view of a simulated image. The synthetic image is generated to mimic real-world conditions, ensuring that the objects appear as they would in an actual captured image. This includes proper lighting, perspective, and spatial relationships between objects. The synthetic image can be used to train or test machine learning models, such as object detection, segmentation, or recognition systems, by providing a controlled yet realistic dataset. The method may also involve adjusting parameters of the synthetic image, such as object positions, orientations, or environmental conditions, to generate diverse training examples. This helps improve the robustness of machine learning models by exposing them to a wide range of scenarios. The synthetic image can be rendered using computer graphics techniques, ensuring high fidelity and accuracy in representing the simulated objects. The approach reduces the reliance on manually annotated real-world data, which can be time-consuming and expensive to obtain.
15. The method of claim 14 , wherein the simulated image is generated by an image generator.
16. The method of claim 15 , wherein the image generator comprises a gaming engine.
17. The method of claim 13 , wherein the synthetic image is generated via a second deep neural network based on the simulated image.
18. The method of claim 17 , wherein the second deep neural network comprises an encoder-decoder neural network and the first deep neural network comprises a convolutional neural network.
19. The method of claim 18 , further comprising: determining that the synthetic image is sourced from the real data distribution when the prediction is greater than or equal to a predetermined threshold.
20. The method of claim 19 , further comprising: storing the synthetic image in a database when the prediction is greater than or equal to the predetermined threshold.
Unknown
April 13, 2021
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.